| Summary: Electricity generation from renewable energy sources is increased day by day and integration of renewable energy sources to the electrical grid leads to some issues in the grid because of intermittent and variable characteristics of these renewable energy sources. Accurate forecasting of electricity generation from the renewable energy sources which have intermittent and variable characteristics is a requirement to ensure stable operation of the electrical grid. Solar irradiation data is also directly associated with photovoltaic (FV) power. FV power forecasting for areas with PV power system studies and without measured PV power data, can be made possible by solar irradiation forecasting. In this thesis, day ahead hourly solar irradiation forecasting has been made to ensure the work of the PV power system in a stable operation and contribute to the load planning work in the smart grid. The next day hourly solar irradiation data for the province of Trabzon have been forecasted by artificial neural network (ANN) approach using one day ahead hourly solar irradiation from the past years. The solar irradiation data is analyzed by a seasonal separation and classified by k-medoids algorithm. As a result of the clustering process, days are divided into three different classes by air type: clear, cloudy and rainy. A different artificial neural network is designed for each class. The designed ANN model has been designed, trained and tested in MATLAB simulation environment without using codes of the MATLAB ANN toolbox. The accuracy of the predictions has been evaluated by different performance measures. |
